CoBELa: Steering Transparent Generation via Concept Bottlenecks on Energy Landscapes

Researchers introduce CoBELa, a new AI framework for interpretable image generation that uses concept bottlenecks on energy landscapes to enable transparent, controllable synthesis without requiring decoder retraining. The system achieves strong performance on benchmark datasets while allowing users to compositionally manipulate concepts through energy function combinations.

• →CoBELa eliminates non-explicit bottleneck representations by conditioning generation entirely through per-concept energy functions over pretrained generator latent spaces.
• →The framework enables compositional concept interventions through additive energy functions, allowing concept conjunction and negation without additional training.
• →A diffusion-scheduled energy guidance scheme replaces expensive MCMC chains with more stable denoising for efficient concept-steered sampling.
• →Experiments show superior performance with 75.70%/82.42% concept accuracy and 6.47/5.37 FID scores on CelebA-HQ and CUB-200-2011 datasets respectively.
• →The decoder-free approach requires no generator retraining and enables post-hoc interpretation of existing models.